Muse reaches beyond its original launch
Meta has described a wider set of uses for Muse, its personal AI agent, following the product’s launch earlier in September. The Meta announcement, dated 24 September, names new shopping and work connectors, voice interactions, a future agent email address and planned access through AI glasses. The subject is the agent’s reach and controls, not the separate Meta VR Glasses hardware launch or Muse Realtime Avatar research.
Meta says Muse already launched with dozens of partners and access to the Shopify catalogue. It is adding retailers including Walmart, Best Buy, Gap, Sephora and Wayfair, alongside Shop Pay and PayPal for payments. Expedia is described as coming soon for travel and Instacart for groceries. The different verbs matter: announced partners and future integrations should not all be treated as live in every account.
For work, Meta names Notion, Granola, GitHub and Box. Those services contain very different kinds of information. A connector that reads meeting notes is not equivalent to one that can edit a repository or share files. Users and administrators should check the exact scopes and action approvals rather than assuming one general Muse permission covers them safely.
Voice while work continues
Meta describes a voice mode intended for long conversations in which Muse can keep working in the background. This is a step beyond a simple question-and-answer exchange: the person may discuss a goal while the agent searches, compares options or performs intermediate steps. That could make complex errands more natural, but it also makes the agent’s current state harder to infer.
A useful voice interface should signal when a task is still running, what has been completed and where approval is required. If a person changes their mind midway through a conversation, Muse should know whether to cancel, revise or finish a previously started action. A spoken “yes” should be tied to a clearly stated proposal, especially before a purchase or message.
The announcement does not supply a universal availability date or a detailed matrix of countries and devices for voice mode. That uncertainty should be preserved. An Australian user should test the actual app and account rather than assume a feature demonstrated at Connect is already enabled locally.
An address for the agent
Meta says Muse will have its own email address, both as a way for a person to contact it and as a means for the agent to get things done. This could help move tasks out of a chat window, but email also carries authority. A message from an agent might be mistaken for one directly written and approved by its owner.
The sensible questions are concrete: can the owner review a draft before it is sent, restrict recipients, see the complete sent history and revoke the address? How will the system handle a malicious instruction arriving by email or embedded in a forwarded document? Meta’s roundup does not answer every question, so a responsible article should not invent defaults.
The company’s initial Muse announcement said users choose whether the agent may only read email or may send on their behalf. The new dedicated-address plan adds another identity to that arrangement. Teams should distinguish the agent’s address from a user’s mailbox, and make it clear to recipients when an automated system is communicating.
Shopping and work need different boundaries
More connectors increase the number of tasks Muse can attempt, but they also create more opportunities for cross-service mistakes. A product recommendation might draw on a shopping history; a payment flow might use an existing account. The owner should be able to tell which service supplied information and which service will receive it before approving a transaction.
Work integrations deserve the same scrutiny. A meeting summary from Granola could be useful in Notion, but not every note should be copied into a wider workspace. A GitHub issue may include confidential code or security details. Testing should include a request the agent is not authorised to complete, and confirm that it stops rather than borrowing access from another connected tool.
Meta also says Muse will come to AI glasses in coming months and may act on what the wearer sees. That contextual input could remove the need to describe a shelf label or flyer, but it raises consent and accuracy issues. A camera-derived interpretation should be checked before an agent purchases, posts or records anything consequential.
A rollout to watch, not flatten
The Connect roundup includes features at different stages: existing connectors, additions being rolled out, services described as coming soon, and hardware integration planned for later. It does not warrant a blanket claim that every listed use is available today. Readers should consult the current Muse interface and partner support pages for their account and country.
The expansion shows Meta’s direction for Muse as an agent that can span shopping, work and communication. That breadth can save effort when tasks cross applications. It also makes authority and traceability more important. The best evaluation will measure completed useful tasks alongside mistaken actions, disclosure and the effort needed for a person to stay in control.